One week / one topic: The kindness of strangers

What happened?

Lately, government bonds and AI stocks have been moving in opposite directions.

Markets are now treating weaker AI spending as a sign of slower growth and less new bond supply, so the AI problem is somehow easing the rates problem.

But is there actually an ‘AI bubble’? And, if so, has it finally started to deflate?

The big names disagree:

• The bears – including Ken Griffin and Ray Dalio – point to bubble-like behaviour, a Shiller CAPE close to its 1999 record and heavy debt-funded spending.

• The bulls – led by Blackstone's Jon Gray – argue that demand is still outrunning capacity, and big tech valuations are nowhere near the excesses of 2000.

While the debate is certainly front and center for most investors, what is not in doubt is the scale of the spending:

• Anthropic's draft IPO prospectus shows revenue of $16.3bn in the first half of 2026, up from $1.1bn a year earlier, and computing costs that could reach $750bn.

• Meta raised its future commitments to about $825bn, from almost $700bn in July, and its quarterly free cash flow turned negative.

Judging from recent price action, investors are starting to push back.

Meta's credit default swaps hit a record, and the strain is even more visible further down the credit ladder.

CCC-rated European (!) spreads have jumped sharply, and some of the largest credit hedge funds suffered heavy losses in the last few weeks.

And let’s not forget, governments are in the same queue for investors' money…

Granted, you can’t really treat government bonds and hyperscaler debt as equivalent – the sovereign entity ultimately makes the rules – but capital is fungible, and to a certain extent the crowding out effect is real.

Right now, France is the most visible example: 10-year yields are above 5% and debt is about 120% of GDP.

Simply put, if the interest rate it pays exceeds nominal growth, stabilizing the debt could require a primary surplus France hasn't run since the 1970s.

However, with protests spreading and an election only months away, political appetite for reform is close to zero… Bonne chance!

What the AI buildout and heavily indebted governments both rely on is the kindness of strangers, and – for now – a bad week for AI makes it a little easier for governments to borrow.

The catch, however, is that this hedge is circular: if AI stabilizes, rates could come straight back under pressure.

How should we manage this in portfolios?

Our observations

• Fundamentals: AI demand isn't in question (yet?), but the money needed to meet it is now very large.

  Governments and hyperscalers are ultimately drawing on the same pool of savings, and neither is planning to borrow less any time soon.

• Price action: The credit sell-off has been fast and concentrated in the weakest borrowers, but it is significant nonetheless.

  Should the weakness continue, I would expect a more meaningful turn in the ‘AGI or die’ narrative…

• Investor beliefs: Is the recent deterioration in sentiment just a bump in the road, or an actual turning point?

  Reaction to announcements from the ongoing corporate earnings season might provide some much-needed clues.

So what?

Those harder questions are now showing up in equity valuations, too.

The Nasdaq 100's forward P/E has fallen to about 20x – from around 28x at its 2025 peak – back near the 2022 lows and not far from where it was before the pandemic.

However, this might not be an obvious buying opportunity. (Do call me if you spot one!)

The de-rating is most likely not just a change in mood...

For a long time, these companies needed little capital, produced enormous amounts of cash and earned near-monopoly profits in their own markets.

Now, they are becoming capital-intensive businesses in a fierce arms race against each other, increasingly funded by the same strangers whose kindness everyone else relies on.

Therefore, a lower P/E for a business with heavier capital needs and more competition isn't obviously a mistake... It may just be the market pricing in a different business model.

Until it's clearer that this spending will earn a decent return on capital, staying selective makes sense and we want to keep our exposure to the overall AI theme at reasonable levels.

Equally, given the circular relationship, we remain skeptical of relying on government bonds alone as a hedge against AI disappointment.

After all, strangers can change their minds with little notice.

Mood music: Bill Withers – Lean on Me

By popular demand, here is the One week / One topic playlist

The information provided should not be considered a recommendation to purchase or sell any particular security.